Table 1. Analysis/Assimilation Differences Possibly Leading to Reduced NAM Performance Relative to GFS


Priority

Difference

Global

Regional

Discussion

Prediction Model

2008/ 2009/ 2010

Prediction model physics/dynamics

GFS

Spectral

NMM

Grid-point

See Table 2

Domain Characteristics

2008

Horizontal Domain & Resolution

Global

22 km

Continental

12 km

Large scales may not be properly corrected

2009/ 2010

Top layer P

Vertical Structure

And Discretization Strategy

0.267 mb

64 layers

Hybrid (sigma-p)

Slowly varying depth (z)

10.9 mb

60 layers

Different Hybrid

Slowly varying mass (p)

While not much mass, there are a lot more layers in GFS’ stratosphere than in NAM’s

Assimilation Cycle Characteristics

2017

Lateral boundary conditions

None

6-hour old GFS

Can fix this with concurrent running of GFS & NAM (can only do this in NEMS + NPOES era ~2017 )

2008/ 2009

Cycles per day / Analysis Updates per cycle / Number - range of forecasts per cycle

4 / 1 / 1 - 9 hr

(extra 3hr of fcst is for FGAT) GDAS does one analysis and one forecast covering 6 hour period

4 / 4 / 4 - 3 hr NDAS does four analyses and four forecasts covering a 12 hour period

NDAS is planned to go to hourly analysis updates to better capture all the available high frequency obs & will probably need DFI

2008

Digital Filter Initialization (DFI)

Yes

No

GFS’ DFI is centered on 3 hr w/ filter width of 6 hr, model is integrated 0-6 hr w/ full physics, then restarted from 3 hr with filtered result.

2007

Precipitation forcing for land-surface energy balance etc.

GFS model precipitation

Stage II/IV precip analysis over CONUS, precip

Considered a strength of NDAS over GDAS

GSI Analysis Characteristics

2007/ 2008

Strong Balance Constraint

Yes

No

A scheme was tried in regional but it assumed f-plane, did not perform well on large NAM domain; trying new approach

2008

Satellite channel bias correction

Global

GFS


Regional

NMM


Can’t impose global sum be zero and can’t use GFS Bias corrections because they depend on model and its vertical structure and discretization

2008/ 2009

First Guess at Appropriate Time (FGAT) / First Order Time-extrapolation to Observation (FOTO)

Yes / Not yet

Not yet / Not yet

Regional update is only 3 hrs so FGAT less necessary, plus FGAT requires more forecast time between analysis updates

2007

Iterations used in variational solution: 1st Outer Loop / 2nd Outer Loop

100 / 150

50 / 50

Regional domain requires fewer iterations to reach same amount of convergence as global

Observational Differences

2017

Data cut-off time

2 hr 45 min

1 hr 15 min

More obs available to GFS due to later data cut off

2009

Observation time window

6 hours

3 hours with plans for hourly to better capture all high frequency obs

Matches the analysis update frequency in GDAS & NDAS and ensures no obs are used more than once

2008

Use of surface land obs of temperature, wind and moisture

No

Yes

Heavily dependent on forward model, vertical background error covariance and its vertical sharpness

2008

Use of 88D radial velocity

No

Yes

Heavily dependent on forward model, vertical background error covariance and its vertical sharpness


Use GPS-IPW

No

Yes

Tests in GDAS had negative impact but positive in NDAS

2009

Use GPS radio occultation refractivity

Yes

Yes


2009

Use of satellite ozone obs

Yes

No

Ozone is a predicted variable in the GFS but not in the NAM

2008

Use METOP-2 HIRS-4, AMSU-A, MHS radiances

Yes

Yes


2007

Use of AQUA/AIRS IR and AMSU-A every field-of-view radiances

Yes

Yes


2009/ 2010

Use of TRMM/TMI and SSM/I rainfall data

Yes

No

Forward model used is based on GFS convective physics

2007

Use GOES-11 and -12 1x1 single field-of-view sounder radiances

Yes over water

Yes over water


2007

Use Aqua/Terra MODIS IR and WV winds

Yes

Yes


2007

Use QuikSCAT scatterometer winds

Yes

Yes


2008

Use of SSM/I wind speed

Yes calculated from Neural Net 3 algorithm

Yes calculated from Goodberlet algorithm (Navy)


2009

Tropical storm relocation

Yes

No


2009

Use of tropical cyclone bogus winds

No (unless weak or new storm and can’t relocate)

Yes always



Table 2. Prediction Model Differences Possibly Leading to Reduced NAM Performance Relative to GFS


Difference

Global

Regional

Discussion

dynamics

Spectral

T574 ~22 km gaussian grid

Grid-point

12 km

Arakawa

B-grid

Can’t run GFS regionally or with nesting, but there is a global version of NMM on a B-grid in new ESMF-based NEMS

Lateral diffusion

Computed on pressure levels & transformed back to hybrid

Computed on hybrid surfaces

A re-formulated lateral diffusion has been coded for NMM

Gravity wave drag (also includes mountain blocking)

Yes


Yes


Vertical diffusion

Yes

MYJ level 2.5 closure in free atmosphere


Land-surface

Noah LSM

13 veg / 9 soil categories

Noah LSM

20 veg / 16 soil categories

terrain height, land-sea mask, roughness length and glacial ice differences exist as well

SST

1 degree Reynolds

0.125 degree RTG_SST

Both are updated once daily

Surface layer

GFS

NMM

Effectively the “marriage” between land or ocean & PBL:

GFS and NAM use different

surface layer schemes

Boundary layer scheme

MRF

Non-local

MYJ

Level 2.5


Shallow Convection

Tiedke

BMJ

GFS’ Tiedke is available in NEMS but only through SAS

Deep Convection

SAS

BMJ


Gridscale clouds, precip microphysics

Zhao

Ferrier

Shortwave Radiation

NASA (simplified, faster Chou)

GFDL (Lacis & Hanson w/ Ferrier tweaks)

Longwave Radiation

RRTM (adapted to GFS from AER version)

GFDL (Fels & Schwartzkopf)